sayzardJonas Hübotter (@jonashuebotter)
Latent Space 팟캐스트(@latentspacepod)에서 Ted Kyi가 진행한 self-distillation(자기증류) 토론을 발견했다는 소식입니다. 발언자는 자기증류에 대한 훌륭한 정리와 향후 연구·적용 가능성에 대한 기대감을 표명하고 있습니다. 연구자·개발자 관점에서 관심을 끌 만한 내용입니다.
Flux | Design PCBs with AI (@BuildWithFlux)
전자제품의 총소유비용(TCO)이 낮아지면서 누구나 자체 하드웨어를 만들기 쉬워졌다는 내용의 하이라이트를 공유한 트윗입니다. 해당 토크는 CEO Matthias Wagner가 @latentspacepod(라텀 스페이스 팟캐스트)에서 발표한 최근 대화에서 발췌된 내용입니다. 하드웨어 접근성 관련 논의입니다.
Flux | Design PCBs with AI (@BuildWithFlux) on X
The total cost of ownership for electronics has made it more accessible for everyone to create their own hardware. Recent highlight from the latest talk with our CEO, @MatthiasWagner , on the @latentspacepod. #hardware #electronics #electronicsengineering #pcbdesign
PsychoticSheepFinally got SDXL running!
I looked into the error
"modules.devices.NansException: A tensor with NaNs was produced in VAE"
and here’s what it means:
Briefly, how an image is generated with a diffusion model: The text encoder interprets the prompt, the UNet "dreams" iteratively in the latent space from noise into an image structure, and the VAE translates this latent vision into visible pixels.
A tensor is simply a multi-dimensional array of numbers, basically the data structure where the model stores all its calculations, like colors, intensities, and intermediate results of the image.
In this case, the VAE experienced a numerical instability: the latent tensor contained invalid values (NaNs), so the dreamed image could not be decoded correctly. In short: the model was still dreaming in the latent space, but the numbers “exploded” along the way (e.g., division by zero, overflow, or undefined operations).
#StableDiffusion #SDXL #AIArt #DiffusionModel #VAE #LatentSpace #Tensor #DigitalArt #CinematicArt #Kunst #KI #AI #DigitalIllustration #StilizedRealism #UrbanFantasy #Motion #DramaticLighting #FilmStill #AICommunity
I looked into the error
"modules.devices.NansException: A tensor with NaNs was produced in VAE"
and here’s what it means:
Briefly, how an image is generated with a diffusion model: The text encoder interprets the prompt, the UNet "dreams" iteratively in the latent space from noise into an image structure, and the VAE translates this latent vision into visible pixels.
A tensor is simply a multi-dimensional array of numbers, basically the data structure where the model stores all its calculations, like colors, intensities, and intermediate results of the image.
In this case, the VAE experienced a numerical instability: the latent tensor contained invalid values (NaNs), so the dreamed image could not be decoded correctly. In short: the model was still dreaming in the latent space, but the numbers “exploded” along the way (e.g., division by zero, overflow, or undefined operations).
#StableDiffusion #SDXL #AIArt #DiffusionModel #VAE #LatentSpace #Tensor #DigitalArt #CinematicArt #Kunst #KI #AI #DigitalIllustration #StilizedRealism #UrbanFantasy #Motion #DramaticLighting #FilmStill #AICommunity
NVIDIA AI Developer (@NVIDIAAIDev)
Nemotron이 Latent Space 팟캐스트에서 언급되어 제작자가 감사 인사를 전했습니다. 이는 Nemotron이라는 모델(또는 프로젝트)이 커뮤니티 미디어에서 언급되며 노출되었음을 알리는 트윗입니다(팟캐스트·미디어 노출 관련 소식).
#BruceSterling
@xuv writesQu’est-ce qu’un espace latent ? Techniquement, il s’agit d’une représentation mathématique intermédiaire, située entre les couches d’entrée et de sortie d’un réseau de neurones, où les données sont encodées sous forme de vecteurs statistiques dans un espace de haute dimensionnalité, des centaines, des milliers, parfois des millions de dimensions. Contrairement aux espaces de représentation traditionnels (pixels pour une image, échantillons pour un son), l’espace latent n’organise pas les données selon des propriétés sensibles immédiatement intelligibles, mais selon des proximités statistiques qui émergent du processus d’apprentissage lui-même.
Petit extrait, ou plutôt définition extraite, du dernier essais “La lisibilité du code” de Grégory Chatonsky. À lire absolument pour comprendre ce qui se passe chez les professionnels du code (les programmeurs) et par, extension, pour nous toustes. Et surtout, ce qu’il va falloir faire pour résister.
https://p.xuv.be/citation-de-gregory-chatonsky-3 #AI #developer #latentSpace
La lisibilité du code / Code readability
On nous répète qu'il faut apprendre à lire le code pour rester libres. Mais les experts eux-mêmes ont cessé de lire celui qu'ils produisent. Plus : les espaces latents des IA résistent structurellement à toute lecture — ce n'est pas un défaut temporaire, c'est leur nature même. L'alphabétisation promise est un leurre. Il nous faut inventer autre chose : non pas comprendre la machine, mais naviguer dans son opacité.
eristhe latent space rover found this
#Ornate #Sculptural #LatentSpace #Colorful #Pattern #Relief #Decorative #Tile #Gradient #Textures #Otherworldly #Stonelike #SciFi #Img2img #AiArt #AiArtists #AiArtCommunity #StableDiffusion
want? have: https://aieris.art/featured/the-latent-space-rover-found-this-eris-and-ai.html
Hacker NewsFirst thoughts on o3 pro
https://www.latent.space/p/o3-pro
#HackerNews #FirstThoughts #o3Pro #TechReview #LatentSpace #Innovation
Simon Dückert#Podcast Tipp im #LatentSpace Podcast: wer sich mal in Tiefe für das Training von LLMs am Beispiel #Llama3 interessiert hört: https://podcasts.apple.com/de/podcast/latent-space-the-ai-engineer-podcast-practitioners/id1674008350?i=1000663106957 - schönes Zitat: "The internet is full of shit" 😂 #GenAI #kimooc24 #loscon24
Latent Space: The AI Engineer Podcast — Practitioners talking LLMs, CodeGen, Agents, Multimodality, AI UX, GPU Infra and al: Llama 2, 3 & 4: Synthetic Data, RLHF, Agents on the path to Open Source AGI auf Apple Podcasts
Sendung Latent Space: The AI Engineer Podcast — Practitioners talking LLMs, CodeGen, Agents, Multimodality, AI UX, GPU Infra and al, Folge Llama 2, 3 & 4: Synthetic Data, RLHF, Agents on the path to Open Source AGI – 23. Juli 2024
HANNΞ MAΞSTravel journal entry 3:
A scenic road carved through a backdrop of artificial beauty. The dual suns gently bowing down, casting a warm golden glow, stretching my shadow as I walked...
Part of the 'Journey Through Latent Space' project: https://www.hannemaes.com/persistent/latentspace/
Created using Midjourney, Procreate and Inkscape.
#art #DigitalArt #ArtistsOnMatsodon #FediArt #MastoArt #b3d #blender #tez #tezos #TezosCollectors #sci-fi #LatentSpace #aiImages #AiAssistedArt
